# Multi-modal Understanding on SEED-IMG

Source: [Multi-modal Understanding on SEED-IMG](<https://www.sota2.com/research/sota/multi-modal-understanding-on-seed-img>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-04-27T07:32:08.213Z

Evaluation of multi-modal understanding capabilities using the SEED-IMG dataset, reporting Accuracy and Relative Accuracy.

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

| Scope | Recorded value |
| --- | --- |
| Task | Multi-modal Understanding |
| Dataset (including any stated split) | SEED-IMG |
| Metric | Accuracy |
| Unit | PERCENT |
| Direction | Higher is better |
| Results loaded | 29 of 29 |
| Coverage | All results currently returned for this benchmark |

Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol.

## Metrics

- [Accuracy](<https://www.sota2.com/research/sota/multi-modal-understanding-on-seed-img.md?metric=Accuracy>): Higher is better; unit: PERCENT.

- [Relative Accuracy](<https://www.sota2.com/research/sota/multi-modal-understanding-on-seed-img.md?metric=Relative%20Accuracy>): Higher is better; unit: PERCENT.

## Recorded results

| Method | Accuracy | Result date | Report paper |
| --- | --- | --- | --- |
| Qwen3-VL-4B | 78.46 | Not recorded | 69eeed6955a3c32af89f9647 |
| LLaVA-OneVision-7B | 76.64 | Not recorded | 69eeed6955a3c32af89f9647 |
| SpB2.0-VL-5B | 76.08 | Not recorded | 69eeed6955a3c32af89f9647 |
| Qwen2.5-VL-3B | 73.88 | Not recorded | 69eeed6955a3c32af89f9647 |
| InternVL2-4B | 73.2 | Not recorded | 69eeed6955a3c32af89f9647 |
| InfiniteVL-4B | 72.9 | Not recorded | 69eeed6955a3c32af89f9647 |
| SmolVLM2-2.2B | 70.69 | Not recorded | 69eeed6955a3c32af89f9647 |
| Vanilla 7B | 69.7 | Not recorded | 68fa760f1e209131cea6b33b |
| VisionZip | 67.5 | Not recorded | 68fa760f1e209131cea6b33b |
| VisionZip | 67 | Not recorded | 68fa760f1e209131cea6b33b |
| LLaVA-1.5 Vanilla 13B | 66.9 | Not recorded | 68fa7ae31e209131cea85b29 |
| VisionZip 🔥 | 66.1 | Not recorded | 68fa7ae31e209131cea85b29 |
| VisionZip | 65.9 | Not recorded | 68fa760f1e209131cea6b33b |
| DivPrune | 65.72 | Not recorded | 68fa7ae31e209131cea85b29 |
| VisionZip | 65.6 | Not recorded | 68fa760f1e209131cea6b33b |
| MaTVLM-3B | 65.6 | Not recorded | 69eeed6955a3c32af89f9647 |
| MMTok | 65.49 | Not recorded | 68fa7ae31e209131cea85b29 |
| VisionZip | 65.2 | Not recorded | 68fa7ae31e209131cea85b29 |
| VisionZip 🔥 | 64.9 | Not recorded | 68fa7ae31e209131cea85b29 |
| MMTok | 64.26 | Not recorded | 68fa7ae31e209131cea85b29 |
| DivPrune | 64.22 | Not recorded | 68fa7ae31e209131cea85b29 |
| VisionZip | 63.8 | Not recorded | 68fa7ae31e209131cea85b29 |
| MMTok | 63.39 | Not recorded | 68fa7ae31e209131cea85b29 |
| Cobra-3B | 63.3 | Not recorded | 69eeed6955a3c32af89f9647 |
| VisionZip | 62.6 | Not recorded | 68fa760f1e209131cea6b33b |
| DivPrune | 62.44 | Not recorded | 68fa7ae31e209131cea85b29 |
| VisionZip | 61.7 | Not recorded | 68fa760f1e209131cea6b33b |
| VisionZip 🔥 | 61.4 | Not recorded | 68fa7ae31e209131cea85b29 |
| VisionZip | 60.4 | Not recorded | 68fa7ae31e209131cea85b29 |

### Qwen3-VL-4B

Accuracy: 78.46. Result ID: 69ef10e047f34ce924d4e9ec.

[Reporting paper](<https://www.sota2.com/research/paper/69eeed6955a3c32af89f9647.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6927db54d6b0dd3342930f66.md>)

Recorded evaluation configuration: {"Model":"Qwen3-VL","Parameter Count":"4B"}. This record does not assert independent reproduction.

### LLaVA-OneVision-7B

Accuracy: 76.64. Result ID: 69ef10e047f34ce924d4e9ee.

[Reporting paper](<https://www.sota2.com/research/paper/69eeed6955a3c32af89f9647.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7c031e209131cea8bdde.md>)

Recorded evaluation configuration: {"Model":"LLaVA-OneVision","Parameter Count":"7B"}. This record does not assert independent reproduction.

### SpB2.0-VL-5B

Accuracy: 76.08. Result ID: 69ef10e047f34ce924d4e9f3.

[Reporting paper](<https://www.sota2.com/research/paper/69eeed6955a3c32af89f9647.md>)

[Method origin paper](<https://www.sota2.com/research/paper/69eeed6955a3c32af89f9647.md>)

Recorded evaluation configuration: {"Model":"SpB2.0-VL","Parameter Count":"5B"}. This record does not assert independent reproduction.

### Qwen2.5-VL-3B

Accuracy: 73.88. Result ID: 69ef10e047f34ce924d4e9eb.

[Reporting paper](<https://www.sota2.com/research/paper/69eeed6955a3c32af89f9647.md>)

Recorded evaluation configuration: {"Model":"Qwen2.5-VL","Parameter Count":"3B"}. This record does not assert independent reproduction.

### InternVL2-4B

Accuracy: 73.2. Result ID: 69ef10e047f34ce924d4e9ed.

[Reporting paper](<https://www.sota2.com/research/paper/69eeed6955a3c32af89f9647.md>)

Recorded evaluation configuration: {"Model":"InternVL2","Parameter Count":"4B","Result Source":"OpenVLM Leaderboard (OpenCompass, 2025)"}. This record does not assert independent reproduction.

### InfiniteVL-4B

Accuracy: 72.9. Result ID: 69ef10e047f34ce924d4e9f2.

[Reporting paper](<https://www.sota2.com/research/paper/69eeed6955a3c32af89f9647.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6938fe986cfd488761a53daf.md>)

[Implementation](<https://github.com/hustvl/InfiniteVL>)

Recorded evaluation configuration: {"Model":"InfiniteVL","Parameter Count":"4B","Result Source":"InfiniteVL (Tao et al., 2025) paper"}. This record does not assert independent reproduction.

### SmolVLM2-2.2B

Accuracy: 70.69. Result ID: 69ef10e047f34ce924d4e9ef.

[Reporting paper](<https://www.sota2.com/research/paper/69eeed6955a3c32af89f9647.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa77d51e209131cea75788.md>)

Recorded evaluation configuration: {"Model":"SmolVLM2","Parameter Count":"2.2B"}. This record does not assert independent reproduction.

### Vanilla 7B

Accuracy: 69.7. Result ID: 69a1699535dfcabb00e6e4dd.

[Reporting paper](<https://www.sota2.com/research/paper/68fa760f1e209131cea6b33b.md>)

Recorded evaluation configuration: {"Base Model":"mini-Gemini 7B","Number of Tokens":"576","Protocol":"Upper Bound"}. This record does not assert independent reproduction.

### VisionZip

Accuracy: 67.5. Result ID: 69a1699535dfcabb00e6e4de.

[Reporting paper](<https://www.sota2.com/research/paper/68fa760f1e209131cea6b33b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa760f1e209131cea6b33b.md>)

[Implementation](<https://github.com/dvlab-research/VisionZip>)

Recorded evaluation configuration: {"Base Model":"mini-Gemini 7B","Number of Tokens":"192"}. This record does not assert independent reproduction.

### VisionZip

Accuracy: 67. Result ID: 69a1699535dfcabb00e6e4df.

[Reporting paper](<https://www.sota2.com/research/paper/68fa760f1e209131cea6b33b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa760f1e209131cea6b33b.md>)

[Implementation](<https://github.com/dvlab-research/VisionZip>)

Recorded evaluation configuration: {"Base Model":"mini-Gemini 7B","Number of Tokens":"192","Fine-tuning Protocol":"Fine-tuned multimodal projector"}. This record does not assert independent reproduction.

### LLaVA-1.5 Vanilla 13B

Accuracy: 66.9. Result ID: 69a877496dbb7075b9508baf.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7ae31e209131cea85b29.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa732e1e209131cea5af59.md>)

[Implementation](<https://llava-vl.github.io/>)

Recorded evaluation configuration: {"Tokens Retained":"576","Compression Ratio":"100%"}. This record does not assert independent reproduction.

### VisionZip 🔥

Accuracy: 66.1. Result ID: 69a877496dbb7075b9508bb2.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7ae31e209131cea85b29.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa760f1e209131cea6b33b.md>)

[Implementation](<https://github.com/dvlab-research/VisionZip>)

Recorded evaluation configuration: {"Tokens Retained":"192","Compression Ratio":"↓ 66.7%"}. This record does not assert independent reproduction.

### VisionZip

Accuracy: 65.9. Result ID: 69a1699535dfcabb00e6e4e1.

[Reporting paper](<https://www.sota2.com/research/paper/68fa760f1e209131cea6b33b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa760f1e209131cea6b33b.md>)

[Implementation](<https://github.com/dvlab-research/VisionZip>)

Recorded evaluation configuration: {"Base Model":"mini-Gemini 7B","Number of Tokens":"128","Fine-tuning Protocol":"Fine-tuned multimodal projector"}. This record does not assert independent reproduction.

### DivPrune

Accuracy: 65.72. Result ID: 69a877496dbb7075b9508bb1.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7ae31e209131cea85b29.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa77c41e209131cea7532a.md>)

Recorded evaluation configuration: {"Tokens Retained":"192","Compression Ratio":"↓ 66.7%"}. This record does not assert independent reproduction.

### VisionZip

Accuracy: 65.6. Result ID: 69a1699535dfcabb00e6e4e0.

[Reporting paper](<https://www.sota2.com/research/paper/68fa760f1e209131cea6b33b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa760f1e209131cea6b33b.md>)

[Implementation](<https://github.com/dvlab-research/VisionZip>)

Recorded evaluation configuration: {"Base Model":"mini-Gemini 7B","Number of Tokens":"128"}. This record does not assert independent reproduction.

### MaTVLM-3B

Accuracy: 65.6. Result ID: 69ef10e047f34ce924d4e9f0.

[Reporting paper](<https://www.sota2.com/research/paper/69eeed6955a3c32af89f9647.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6938fe986cfd488761a53daf.md>)

[Implementation](<https://github.com/hustvl/InfiniteVL>)

Recorded evaluation configuration: {"Model":"MaTVLM","Parameter Count":"3B","Result Source":"InfiniteVL (Tao et al., 2025) paper"}. This record does not assert independent reproduction.

### MMTok

Accuracy: 65.49. Result ID: 69a877496dbb7075b9508bb3.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7ae31e209131cea85b29.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7ae31e209131cea85b29.md>)

Recorded evaluation configuration: {"Tokens Retained":"192","Compression Ratio":"↓ 66.7%"}. This record does not assert independent reproduction.

### VisionZip

Accuracy: 65.2. Result ID: 69a877496dbb7075b9508bb0.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7ae31e209131cea85b29.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa760f1e209131cea6b33b.md>)

[Implementation](<https://github.com/dvlab-research/VisionZip>)

Recorded evaluation configuration: {"Tokens Retained":"192","Compression Ratio":"↓ 66.7%"}. This record does not assert independent reproduction.

### VisionZip 🔥

Accuracy: 64.9. Result ID: 69a877496dbb7075b9508bb6.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7ae31e209131cea85b29.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa760f1e209131cea6b33b.md>)

[Implementation](<https://github.com/dvlab-research/VisionZip>)

Recorded evaluation configuration: {"Tokens Retained":"128","Compression Ratio":"↓ 77.8%"}. This record does not assert independent reproduction.

### MMTok

Accuracy: 64.26. Result ID: 69a877496dbb7075b9508bb7.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7ae31e209131cea85b29.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7ae31e209131cea85b29.md>)

Recorded evaluation configuration: {"Tokens Retained":"128","Compression Ratio":"↓ 77.8%"}. This record does not assert independent reproduction.

### DivPrune

Accuracy: 64.22. Result ID: 69a877496dbb7075b9508bb5.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7ae31e209131cea85b29.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa77c41e209131cea7532a.md>)

Recorded evaluation configuration: {"Tokens Retained":"128","Compression Ratio":"↓ 77.8%"}. This record does not assert independent reproduction.

### VisionZip

Accuracy: 63.8. Result ID: 69a877496dbb7075b9508bb4.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7ae31e209131cea85b29.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa760f1e209131cea6b33b.md>)

[Implementation](<https://github.com/dvlab-research/VisionZip>)

Recorded evaluation configuration: {"Tokens Retained":"128","Compression Ratio":"↓ 77.8%"}. This record does not assert independent reproduction.

### MMTok

Accuracy: 63.39. Result ID: 69a877496dbb7075b9508bbb.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7ae31e209131cea85b29.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa7ae31e209131cea85b29.md>)

Recorded evaluation configuration: {"Tokens Retained":"64","Compression Ratio":"↓ 88.9%"}. This record does not assert independent reproduction.

### Cobra-3B

Accuracy: 63.3. Result ID: 69ef10e047f34ce924d4e9f1.

[Reporting paper](<https://www.sota2.com/research/paper/69eeed6955a3c32af89f9647.md>)

[Method origin paper](<https://www.sota2.com/research/paper/6938fe986cfd488761a53daf.md>)

[Implementation](<https://github.com/hustvl/InfiniteVL>)

Recorded evaluation configuration: {"Model":"Cobra","Parameter Count":"3B","Result Source":"InfiniteVL (Tao et al., 2025) paper"}. This record does not assert independent reproduction.

### VisionZip

Accuracy: 62.6. Result ID: 69a1699535dfcabb00e6e4e3.

[Reporting paper](<https://www.sota2.com/research/paper/68fa760f1e209131cea6b33b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa760f1e209131cea6b33b.md>)

[Implementation](<https://github.com/dvlab-research/VisionZip>)

Recorded evaluation configuration: {"Base Model":"mini-Gemini 7B","Number of Tokens":"64","Fine-tuning Protocol":"Fine-tuned multimodal projector"}. This record does not assert independent reproduction.

### DivPrune

Accuracy: 62.44. Result ID: 69a877496dbb7075b9508bb9.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7ae31e209131cea85b29.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa77c41e209131cea7532a.md>)

Recorded evaluation configuration: {"Tokens Retained":"64","Compression Ratio":"↓ 88.9%"}. This record does not assert independent reproduction.

### VisionZip

Accuracy: 61.7. Result ID: 69a1699535dfcabb00e6e4e2.

[Reporting paper](<https://www.sota2.com/research/paper/68fa760f1e209131cea6b33b.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa760f1e209131cea6b33b.md>)

[Implementation](<https://github.com/dvlab-research/VisionZip>)

Recorded evaluation configuration: {"Base Model":"mini-Gemini 7B","Number of Tokens":"64"}. This record does not assert independent reproduction.

### VisionZip 🔥

Accuracy: 61.4. Result ID: 69a877496dbb7075b9508bba.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7ae31e209131cea85b29.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa760f1e209131cea6b33b.md>)

[Implementation](<https://github.com/dvlab-research/VisionZip>)

Recorded evaluation configuration: {"Tokens Retained":"64","Compression Ratio":"↓ 88.9%"}. This record does not assert independent reproduction.

### VisionZip

Accuracy: 60.4. Result ID: 69a877496dbb7075b9508bb8.

[Reporting paper](<https://www.sota2.com/research/paper/68fa7ae31e209131cea85b29.md>)

[Method origin paper](<https://www.sota2.com/research/paper/68fa760f1e209131cea6b33b.md>)

[Implementation](<https://github.com/dvlab-research/VisionZip>)

Recorded evaluation configuration: {"Tokens Retained":"64","Compression Ratio":"↓ 88.9%"}. This record does not assert independent reproduction.

Page 1 of 1. This is an index page, not the full collection.
